The relationship between the scaled selection coefficient, <i>γ</i>, and the dN/dS ratio as estimated by the PAML package from simulated data.

Wright-Fisher simulations based on the full genetic code were performed as described for two independent populations (middle column) and a single population (right column). The table shows the most-likely dN/dS value as estimated by PAML for two sampled sequences, as well as a 95% confidence interval obtained from the χ2 distribution. For each value of γ, the first line corresponds to simulations with μ = 10−7, and the second line corresponds to simulations with μ = 10−6.